Mobile Collateral versus Immobile Collateral

Mobile Collateral versus Immobile Collateral∗

Gary Gorton, Yale and NBER

Tyler Muir, Yale

April 27, 2015

Abstract

The financial architecture prior to the recent financial crisis was asystem of mobile collateral. Safe debt, whether government bondsor privately-produced bonds, i.e., asset-backed securities, could betraded, posted as collateral, and rehypothecated, moving to its high-est value use. Since the financial crisis, regulatory changes to thefinancial architecture have aimed to make collateral immobile, mostnotably with the BIS “liquidity coverage ratio” for banks. We evalu-ate this immobile capital system with reference to a previous regimewhich had this feature: the U.S. National Era.

∗This paper was prepared for the June 2015 BIS Annual Conference in Lucerne, Switzer-land. Thanks to Adam Ashcraft, Darrell Duffie, Arvind Krishnamurthy, and seminarparticpants at the Stanford Global Crossroads Conference for comments. Thanks toCharles Calomiris, Ben Chabot, Michael Fleming, Ken Garbade, Joe Haubrich, JohnJames, Richard Sylla, Ellis Tallman, Warren Weber and Rosalind Wiggins for answer-ing questions about data. Thanks to Lei Xie and Bruce Champ (deceased) for sharingdata. Thanks to Toomas Laarits, Rhona Ceppos, Ashley Garand, Leigh-Anne Clark andMichelle Pavlik for research assistance. Special thanks to the Federal Reserve Bank ofCleveland for sharing the data of the late Bruce Champ.

1 Introduction

In the thirty years prior to the 2007-2008 financial crisis, the global financialsystem evolved away from a system of immobile collateral into a systemof mobile collateral. In this new system bank loans, instead of remainingpassively immobile on bank balance sheets, were securitized into bonds whichcould be traded, used as collateral in repo, posted as collateral for derivativespositions, and rehypothecated, moving to the location of their highest valueuse. In short, U.S. banks’ loans, which had been sitting passively on bankbalance sheets, were transformed into bonds, making them mobile. Since thefinancial crisis of 2007-2008, regulatory initiatives have been aimed at makingcollateral once again immobile. In this paper, we evaluate the recent policyof again making collateral immobile. We have two main arguments in ourevaluation. The first is that there is a cost to making this collateral immobilebecause it ties up safe debt. The second is that we find weak support forthe idea that making collateral immobile has large benefits – namely that itmakes the financial system safer and reduces panics. Instead we find thatother forms of bank debt increase when safe collateral becomes immobile,possibly making the system riskier.

The transformation of the U.S. financial system towards mobile collateralis shown in Figure 1 (from Gorton, Lewellen and Metrick (2012)). The fig-ure shows the different forms of privately-produced safe debt as a percentageof total privately-produced safe debt. In the 1950s and 1960s demand de-posits of banks were 80 percent of the privately-produced safe debt. Demanddeposits were backed by bank loans, essentially the collateral. Demand de-posits then go into a downward trend for the next thirty years. At the sametime, the categories of privately-produced safe debt that are growing weremoney market instruments (specifically, repo, commercial paper and moneymarket funds), as well as AAA mortgage-backed securities (MBS) and otherAAA asset-backed securities (ABS). By the start of the financial crisis, this“shadow banking system” was as large as the amount of demand deposits(though much of the ABS and MBS ended up as collateral backing repo andasset-backed commercial paper). This shadow banking system is funding thevery same bank loans in a different way.

At the same time that this transformation of the financial system wasgoing on, another important, related, trend was developing. This is shownin Figure 2. Figure 2 shows the holders of outstanding U.S. Treasury debt(excluding Treasuries held by the federal, state and local pensions; social se-

1

curity is also not counted). Again, there is a steady downward trend for U.S.banks, as the banks no longer need Treasuries as a component of their port-folios backing demand deposits. Coincidentally, the Rest of the World showsa sharply upward trend in holdings of U.S. Treasury debt. The Rest of theWorld has a demand for safe, liquid, debt. In other words, Treasury debt hasa convenience yield. Krishnamurthy and Vissing-Jorgensen (2012a,b) showempirically that investors value the money-like properties of liquidity andsafety of these bonds. Furthermore there is a negative correlation betweenoutstanding U.S. Treasury debt and the production of privately-producedsafe debt (see Krishnamurthy and Vissing-Jorgensen (2012a,b) and Gorton,Lewellen and Metrick (2012)). When Treasuries outstanding (as a percent-age of GDP) declines, privately-produced safe debt increases to fill the gap.More specifically, Xie (2012) shows, using daily data, that asset-backed secu-rity (ABS) and mortgage-backed security (MBS) issuance occurs when thereis a high convenience yield on Treasuries. In other words, when Treasuries arescarce, more ABS and MBS are issued (85 percent of an issue is AAA). Sun-deram (2014) shows that a high convenience yield results in the endogenousresponse of more asset-backed commercial paper (ABCP) being issued.

The private response to create more privately-produced safe debt whenthere is a scarcity of U.S. Treasuries is partly the focus of the literatureon the “global savings glut”. This literature argues that increased capitalflows into the United States from countries with an excess of savings mayhave been an important reason that U.S. interest rates were low (see, e.g.,Bernanke (2005, 2007), Bernanke, Bertaut, De Marco and Kamin (2011)).Foreign investors have a demand for safe assets and focused their investmentson U.S. Treasuries (see Caballero and Krishnamurthy (2009)). Maggiori(2013) argues that this as a form of insurance; the U.S. can (almost uniquely)produce safe debt, which is demanded by foreign investors in countries whichcannot produce safe debt (China and the oil-producing countries; see Bertaut,De Marco and Kamin (2011)).

The creation of privately-produced safe debt is in part a response toa scarcity of government-produced safe debt. Both types of safe debt—government-produced and privately-produced– are used as collateral for short-term bank money, i.e., repo, ABCP and money market funds (MMF). Theresponse to scarcity was not only increased private production of safe debt,but also increased mobility of the debt. In the past, when U.S. Treasuries orbank loans backed demand deposits, the backing loans were immobile. In-deed, since demand deposits were the dominant form of inside money, there

2

was no need for mobility. The bank loans sat on bank balance sheets andwere not traded. As the need for demand deposits receded, and foreigninvestors demanded U.S. Treasuries, the financial system transformed theimmobile bank loan collateral into forms of mobile collateral via securitiza-tion. Privately-produced safe debt became a product to be used as mobilecollateral.

In Section 2 of this paper we provide evidence of the extent to which thefinancial system became a system of mobile collateral, indeed, stretching theavailable mobile debt to meet demands. To do this we study the determinantsand extent of repo fails. We show that repo fails were increasing because ofthe scarcity of U.S. Treasuries and Agency bonds. Another manifestationof system morphing was that repo was significantly expanding beyond theprimary dealers prior to the crisis. And, fails were caused in significantpart by a demand for liquidity in a world with insufficient safe debt. Thiswas, and remains, a problem. Krishnamurthy and Vissing-Jorgensen (2012b)argue empirically that when the ratio of privately-produced safe debt goes uprelative to Treasuries, financial crises are more likely. Also see Gourinchasand Jeanne (2012). This scarcity appears to persist today. For example,Bertaut, Tabova and Wong (2014) show that since the financial crisis U.S.investors have invested in sovereign debt of Australia and Canada becauseof the shortage of safe debt since securitization collapsed.

New regulations do not address potential scarcities of safe debt. In fact,since the financial crisis, new regulations aim at returning to a financialsystem of immobile collateral. For example, under Dodd-Frank and simi-lar European legislation collateral must be posted to central clearing parties(CCP) (regardless of the private party’s net position), while the CCP doesnot post collateral to participants. CCPs will only accept highly liquid, highgrade collateral. Variation margin has long been part of the bilateral swapmarket, but importantly, initial margin is new and will increase substan-tially the amount of collateral required. Not all swaps trades will be clearedthrough a CCP. For those that are not, initial and variation margin for eachtrade must be held by a third party. Further, collateral posted to banks byclients cannot be rehypothecated. And, most importantly, the BIS LiquidityCoverage Ratio (LCR) requires that (net) short-term bank debt be backedby (essentially) Treasuries (“high-quality liquid assets”).1 See BIS (2013).

1Throughout the paper, we will refer to ”high-quality liquid assets” as Treasuries andwe will refer to net short term debt as repo.

3

In other words, short-term bank money (repo and CP) and Treasuries, bothof which have a convenience yield, must be combined. One kind of moneymust be backed by another kind of money, a kind of narrow banking.

The LCR is the leading example of the move back towards an immobilecollateral system. In effect, it attempts to reverse thirty years of change. Assuch an important change, how should we evaluate the LCR (or any newpolicy for that matter)? Because of the Lucas (1976) critique, a generalequilibrium model would be needed, one which currently does not exist.There have, of course, been numerous more or less ad hoc forecasts of howmuch collateral the new system will need given the LCR, but these numbersvary a lot and are subject to the Lucas critique, i.e., there is no accounting forthe general equilibrium effects that might occur (”unintended consequenfes”).In fact, we know from the above studies that privately-produced money growswhen there is a scarcity of Treasuries.

We are not interested here in forecasting the amount of collateral needed.2

Rather, in this paper we evaluate the LCR structure by analyzing a financialsystem that had the same structure of requiring that U.S. Treasuries backthe issuance of privately-issued bank money: the U.S. National Banking Era,1863-1914.3Under the National Banking System, national banks could issuedistinct “national bank notes” by depositing eligible U.S. Treasury bondswith the U.S. Treasury, which would then print the bank’s notes. Originally,the idea was to create a demand for U.S. Treasuries so as to finance the U.S.Civil War. But, a by-product did not go unnoticed, namely, the belief thatbacking private money with Treasuries would prevent banking panics. Priorto the National Banking Era, U.S. banks issued their own distinct notes,backed by state bonds (in Free Banking states) or backed by portfolios ofbank loans (in chartered banking states). There were systemic banking crisesin 1814, 1819, 1837 and 1857. It was expected that the National BankingSystem would eliminate panics. Similarly, the explicit purpose of the liquiditycoverage ratio (LCR) is to make the financial system safer. BIS (2013) labelsthe LCR as one of the key reforms ”to develop a more resilient banking sctor”(p. 1).

However, this stability did not occur under the National Banking Era.Banking panics were not prevented, but merely shifted from one form of bank

2On this topic see Heller and Vause (2012), Sidanius and Zikes (2012), Fender andLewrick (2013) and Duffie, Scheicher and Vuillemey (2014).

3On the National Banking Era see Noyes (1910), Friedman and Schwartz (1963), andChamp (2011c).

4

money to another. During a panic, instead of requesting cash for privatebank notes, debt holders demanded national bank notes for their demanddeposits. By the time the National Bank Acts were passed demand depositshad become a sizeable form of privately-produced bank money and we showthat they continued to grow substantially throughout this period (Figure9). But, economists did not understand this for decades. Bray Hammond(1957), in his Pulitzer Prize-winning book Banks and Politics in America,wrote: “. . . the importance of deposits was not realized by most Americaneconomists . . . till after 1900” (p. 80). Hammond goes on to discuss why thegrowing importance of demand deposits was overlooked. Economist CharlesDunbar (1887) wrote in the inaugural volume of the Quarterly Journal ofEconomics : “The ease with which we ignore deposits as a part of the currencyseems the more remarkable, when we consider that few men in businessfail to recognize the true meaning of this form of bank liability” (p. 402).And Russell C. Leffingwell, the Assistant Secretary of the Treasury wroteas late as 1919: “All of these people who believe in the quantity theory ofmoney . . . choose to call bank deposits money, but bank deposits are notmoney” (Leffingwell letters, quoted by Wicker (1966, p. 21). Regulatorsand economists were conceptually confused.What seems so obvious now wasnot obvious then. Forms of short-term bank debt change and this was notimmediately recognized. Worse, the new forms of bank debt and their riskstend to be misunderstood. It is possible that the system even contributedto panics by encouraging a different, less well understood form of bank debtto be created. This is a key point because a main benefit of the proposedsystem of immobile collateral is that it will make the system safer.

There was another problem with the National Banking System: too littlemoney was issued and the system was inflexible or “inelastic”. Too littlemoney was issued even though it was apparently profitable to do so, an ap-parent riskless arbitrage opportunity. Economists have called this the “underissuance puzzle” or the “national bank note puzzle.” First noticed by Bell(1912), this under issuance has been a puzzle ever since, for over a century!The puzzle is that national banks never fully utilized their note-issuing pow-ers even though it appears that it was profitable to do so. As Kuehlwein(1992) put it: “. . . through the turn of the century and into the 1920sbanks devoted a significant fraction of their capital to direct loans . . . de-spite the fact that national bank notes appeared to be more profitable” (p.111). Friedman and Schwartz (1963, p. 23) reached the same conclusion.There is a large literature on this; see Calomiris and Mason (2008) and the

5

citations therein.Because the LCR is structurally the same as the National Banking Sys-

tem, this puzzle is important. In this paper, we show that the reason that“riskless arbitrage profits” persisted during the National Banking Era wasthat the calculations of the arbitrage profit done to date ignored the factthat there was a convenience yield to Treasuries and a cost to bank capital.Banks held Treasuries on their balance sheets but, in principle, could haveraised capital to buy Treasuries.4 But, for the system as a whole, there ap-pears to have been a shortage of safe debt. Simply put, banks had otherimportant uses for Treasuries and bank capital was expensive. We show thatthe “arbitrage profits” are essentially a proxy for the “convenience yield” onTreasuries or the cost of bank capital or likely both. This suggests that back-ing one kind of money (National Bank notes) with another kind of money(Treasuries) may not be such a good idea. By linking the two forms of money,another form of private-produced money is likely to appear or grow. Thisis strongly shown in the data – as the share of Treasuries to GDP declinedover this period, deposits grew. And a shortage of safe debt is associatedwith financial instability. This too is consistent with the data as bankingpanics occurred frequently throughout the period (1873, 1884, 1890, 1893,1896, and 1907).

Of course, it will be objected that the two systems—the current LCRand the historical National Banking Era– are different and that the NationalBanking Era is not relevant. Indeed, there are important differences. Never-theless, it seems useful to analyze a system which is structurally identical tothe LCR. In terms of policy evaluation, there seems no real alternative.

We summarize with the following conclusions about the LCR or othersystems that make collateral immobile. First, these systems have costs be-cause they tie up safe assets and add to issues of scarcity. Second, we castdoubt on the systems supposed benefits – namely preventing crises. Histori-cally, the National Banking Era had many panics. It is likely that a systemof immobile collateral which restricts certain forms of bank debt creationsimply encourages other forms of bank debt to be created. This is consistentwith the growth in deposits during the National Banking Era. There is aremarkably strong correlation between Treasury supply and deposits (Figure9). These new forms of bank debt are dangerous because they are typically

4Also, average profit rather than marginal profit was calculated. This distinction isimportant, as we discuss below.

6

not well understood or acknowledged, as deposits were not well understoodin the National Banking Era and as shadow money (repo, ABCP, etc.) werenot understood until the recent crisis.

The paper proceeds as follows. In Section 2 we provide more evidenceon the scarcity of safe assets in the period leading up to the financial crisisof 2007-2008 by looking at repo fails. Section 3 is devoted to the NationalBanking Era. We calculate the profitability of national bank note issuanceand then show that even in the 19th and early 20th centuries U.S. Treasurieshad a convenience yield. We then show that this can resolve the century-oldnational bank note paradox. Section 4 concludes with implications for thepresent day.

2 Collateral Mobility and Scarcity

Many authors have discussed the shortage of safe debt prior to the financialcrisis, e.g., Caballero (2010), Gourinchas and Jeanne (2012), usually relatingit to the global savings glut. But, it has proven difficult to provide evidencefor this shortage. In this section we provide some evidence for this shortage,which was driving the growth of mobile collateral.

2.1 Repo Fails

There is said to be a “repo fail” if one side to the repo transaction doesnot abide by the contract at maturity, failing to deliver the collateral back(called a “failure to deliver”) or failing to repay the loan (called a “failureto receive”). See Fleming and Garbade (2005, 2002) on fails. Repo failscan provide indirect evidence on scarcity and mobility. If collateral is scarce,then it can become more mobile via rehypothecation (re-use) chains, makingit more difficult to find the bond to return to the borrower, i.e. a fail. Thereis no direct evidence on this, but we provide a variety of indirect evidence.

We examine data from the New York Federal Reserve Bank on primarydealers’ fails.5 The primary dealers are only a subset of all firms involvedin the bilateral repo market, as we will see below. But, still it encompassesmany large financial firms. The New York Federal Reserve Bank collects

5“Primary dealers” are financial firms that are trading counterparties if the New YorkFed in its implementation of monetary policy. There are currently 22 primary dealers; seehttp://www.newyorkfed.org/markets/pridealers current.html .

7

data on only three asset classes used as collateral for repo: U.S. Treasuries,Agency bonds, and Agency MBS.6

Repo fails by asset class are shown in the three panels of Figure 3. FromFigure 3 it is apparent that repo fails were increasing prior to the financialcrisis. It is apparent from the figure that the period from January 2000 untilJanuary 2010 is more turbulent than the period before and the period after.The turbulence is not just the financial crisis. This is confirmed by Table1 Panel A which shows the mean dollar amount of fails (in $ millions) inthe 1990s compared to the period 2000-2007; also shown are the standarddeviation of fails. We formally test for difference between subperiods below.

The data collected by the New York Fed is very limited. To get some senseof the narrowness of the primary dealer group, we can look at data from theDepository Trust and Clearing Corporation (DTCC) on fails. DTCC hashundreds of members that use DTCC for clearing and settlement.7 In 2011DTCC settled $1.7 quadrillion in security value. DTCC also has a large repoprogram. DTCC fails data is for the value of Treasury and Agency fails, thatis the amounts that were not delivered to fulfill a contract. The DTCC datacovers all fails of Treasuries and Agencies, not just repo fails. However, ifthere is a scarcity of safe debt, then there are likely fails in trades as wellas repo. The DTCC series is not as long as the NY Fed’s, but it shows thelarger universe of players. The data are shown in Figure 4; looking at thescale of the y-axis it is clear that there are many more fails, suggesting thatthe size of the fails problem is an order of magnitude larger than the NY Feddata shows. Also, see Gorton and Metrick (2015).

Aside from operational issues that explain repo fails, there are two otherpossibilities. First, there is the possibility that a counterparty strategicallydefaults to retain the bonds or retain the cash, at least for a short periodof time. Secondly, there can be multiple fails due to rehypothecation (there-use of collateral) chains, i.e., several transactions are sequentially basedon the same collateral. As explained by Fleming and Garbade (2002): “ . .. a seller may be unable to deliver securities because of a failure to receivethe same securities in settlement of an unrelated purchase. This can lead toa ‘daisy chain’ of cumulatively additive fails: A’s failure to deliver bonds toB causes B to fail on a sale of the same bonds to C, causing C to fail on a

6“Agency” refers to Fannie Mae, Freddie Mac or Ginnie Mae, government-sponsoredenterprises that securitize and guarantee certain types of residential mortgages.

7See the DTCC membership list: http://www.dtcc.com/client-center/dtc-directories.aspx .

8

similar sale to D, and so on” (p. 43). We do not have the data, however, todistinguish between fails due to rehypothecation chains from other fails. Wecannot distinguish between these possibilities, but the tests below stronglysuggest that increasingly fails were not operational errors.

Collateral is mobile if it is in a form that can be traded and posted ascollateral in repo or derivatives transactions. Rehypothecation is anotherform of collateral mobility. What is the extent of rehypothecation? There issome survey data from the International Swaps and Derivatives Association(ISDA). ISDA has an annual survey of its members that usually asks aboutthe extent of rehypothecation using collateral received in OTC derivativetransactions, in terms of the percentage of institutions that report that theydo rehypothecate collateral. In 2001, the first survey, 70 percent or therespondents reported that they “. . . actively re-use (or ‘rehypothecate’)incoming collateral assets in order to satisfy their own outgoing collateralobligations” (p. 3). Over the years the percentage rises to 96 percent forlarge firms in 2011. In 2014, ISDA for the first time asked about whichbonds were actually used for rehypothecation. Table 1 Panel C shows theresults. In 2014 ISDA estimated that total collateral used in non-clearedOTC derivatives to be $3.7 trillion. It would appear that rehypothecation issizeable. This does not address the question of the length of rehypothecationchains. Singh (2011) estimates that prior to the financial crisis, collateralvelocity was three. Also see Singh and Aitken (2010).

This is not the only evidence on scarcity and mobility. The bilateral repomarket was expanding significantly beyond the primary dealers in the 2000s.In the New York Fed data, if one dealer fails to deliver to another dealer,then the first dealer records a “fail to deliver” of $N, and the counterpartyprimary dealer reports a “fail to receive” of $N. So, fails and receives shouldbe equal, unless the primary dealers are trading with firms that are notprimary dealers.

To examine whether repo was expanding beyond the primary dealers welook at the difference between receive and fail by asset class. If all the fails arebetween primary dealers, then this number will be zero. So, if this numberis positive, then it means that the party failing to deliver was not a primarydealer, the primary dealer records a “fail to receive”. Figure 5 shows failureto receive minus failure to deliver by asset class. Again it is apparent that thisnumber was near zero prior to 2000, meaning that all fails were with anotherprimary dealer. But, after 2000 and prior to the crisis, Receive minus Deliveris clearly not zero. In this period there are significant fails by non-primary

9

dealer counterparties, suggesting that the bilateral repo market had grownsignificantly, consistent with collateral being mobile and scarce. (Also seeGorton and Metrick (2015).

This is confirmed in Table 1 Panel B, where it is clear that failure toreceive minus failure to deliver increasingly differs from zero in the period2000-2007, prior to the crisis. Moreover, note the sign difference betweenTreasuries and MBS during 2000-2007. For Treasuries receive minus fail isvery large in 2000-2007, again meaning that non-primary dealers are notdelivering Treasuries according to their repo contracts. But, in the case ofMBS, the number is very negative, meaning that primary dealers are failingto deliver to non-primary dealer counterparties. This is also apparent in thefigure.

The fails data on Agency MBS market is very different, likely becausethe repo fails number includes fails in the “to be announced” (TBA) market,although the data do not allow us to decompose the fails. (See GovernmentSecurities Dealers Reports (2015), Board of Governors of the Federal ReserveSystem). TBA contracts are forward contracts for the purchase of “to beannounced” agency MBS. In this market, the MBS to be traded are notspecified initially. Rather, the parties agree on six general parameters of theMBS (date, issuer, interest rate, maturity, face amount, price). The contractsinvolve a delayed delivery, typically an interval of several weeks. In theTBA market transactions are usually “dollar rolls,” agreements involving thepurchase or sale of an agency MBS with a simultaneous agreement to resellor repurchase MBS at a specified price. In a TBA dollar roll the securitiesreturned need only be “substantially similar,” unlike in a repo transaction.The TBA market is very large. Average daily fails in this market betweenDecember 31, 2009 and December 29, 2010, as reported by primary dealers,was $83.3 billion in fails to deliver and $73.8 billion in fails to receive (seeTreasury Market Practices Group (TMPG) (2011)). On the TBA market,see Vickery and Wright (2013).

2.2 Fails and the Demand for Liquidity

We now turn to some formal evidence that collateral became increasinglymobile. We start the analysis by testing to see if there are significant break-points in the panel of fails (receive and deliver) data. To do this we followBai (2010). Bai (2010) shows how to find breakpoints in panels of datawhere a breakpoint is in the mean and/or the variance. Assuming a com-

10

mon breakpoint in a panel of data is more restrictive than assuming randombreakpoints in the individual different series in the panel, but the methodcan be used on an individual series as well.

The method can be used to find other breakpoints subsequent to the first.The first breakpoint divides the panel into two sub-panels, on each side ofthe first breakpoint. To find the second breakpoint apply the procedure toeach of the two subseries, on the two sides of the first breakpoint. The secondbreakpoint is the one that gives the larger reduction in the sum of squaredresiduals, when comparing the break found in each of the two subseries.

We examine a panel of four series: fail to deliver and fail to receive forTreasuries and for Agencies. We omit MBS for reasons discussed above.The sample period of weekly data runs from July 1991 to September 2014.The breakpoints are shown in Table 2. The table shows the 95% confidenceintervals in the last two columns in terms of dates. From the figures aboveit is clear that fails are increasing, starting in the early 2000s. Consistentwith this the first breakpoint is September 12, 2001, just after September11, 2001. This is the start of a different regime and it extends until, notsurprisingly, the second break chronologically just after Lehman. The thirdbreakpoint is February 9, 2009.

Why were fails increasing? We will examine the proposition that failsincreased as the demand for liquidity increased. We follow Xie (2012) inmeasuring the convenience yield by the spread between the rate on generalcollateral (GC) repo and the rate on the Treasury used as collateral for therepo. The maturity is one month. In GC repo, the lenders will accept anyof a variety of Treasuries as collateral, i.e., it is general collateral rather thanspecific collateral. Xie shows that this spread has historically (1991-2007)been 36 basis points (see Xie (2012)), reflecting the fact that with GC repothe Treasury must be returned and the cash is tied up during the time of therepo Obtaining a Treasury via reverse repo is not as good as actual ownershipof the Treasury.

To be clear, the GC repo to Treasury spread of 36 basis points meansthat the borrower is losing money to the lender for the cash lent. Takethe more usual case, say with privately-produced collateral. Then, if thisunderlying collateral earned 6 percent and the repo rate was 3 percent, thenthe borrower would be earning the spread of six minus three. But withTreasuries as collateral, because of their safety and liquidity, the borrower isearning negative 36 basis points. The borrower is paying to own a Treasuryfinanced in the repo market. The amount of repo supplied then is linked to

11

the convenience yield of Treasuries.The basic idea we explore is whether an increase in the GC repo spread,

i.e., an increase in the convenience yield, is associated with an increase inrepo fails. In other words, if there is an increase in the demand for liquidity,then this spread will widen. A widening of the spread corresponds to anincreased scarcity of Treasuries, and possibly other safe collateral as well,such as Agency bonds and Agency MBS.

We use differences-in-differences in seemingly-unrelated regression on thepanel of the Treasuries and Agency bonds, where fails are normalized byfails in 2013. We indicate the three breaks discussed above. The first periodis September 12, 2001 going until September 23, 2008 and the second isSeptember 24, 2008 until February 10, 2009, followed by February 11, 2009onwards. This means that there are four periods: prior to 2001, break 1,break 2 and break 3. We will look at specifications with and without lags.We also include the change in the one month T-bill since the level of theinterest rate effects the incentive to fail. In a repo fail the implicit penalty isthe interest that could have been earned elsewhere, so in a low interest rateenvironment the penalty is low.8

The regression results for fails to receive are shown in Table 3. The tablefor fails to deliver is in the Appendix. In both cases the interaction betweenthe GC repo spread and regime 1 is significant. Break 1 is the period priorto the crisis up to September 24, 2008. In both cases the interaction of thespread with the period of Break 2 is also significant and these coefficientsare larger. Break 2 runs from September 24, 2008 to February 11, 2009, thepost-Lehman period. Changes in the convenience yield or the demand forliquidity appear to have driven repo fails in the period prior to the financialcrisis and during the crisis. The first regime corresponds to the period ofthe scarcity of safe debt while the second regime corresponds to the flight toquality. This is true for both fails to receive and fails to deliver. The changein the one month T-bill rate is also significant, suggesting that the incentiveto fail is related to the level of the interest rate. Finally, the three dummyvariables for the three break regimes are not significant. This may be due toa combination of factors. The break points may be driven by the variance,and the interaction terms may be absorbing this effect.

8For this reason the Treasury Market Practices Group introduced a “dynamic failscharge” to provide an incentive for timely settlement. See Garbade, Keane, Logan Stokesand Wolgemuth (2010).

12

In the Appendix we examine the breakpoints for the absolute value offails to deliver minus fails to receive in the Treasury and Agency MBS repomarkets. This is the variable that measures the growth of the repo marketbeyond the primary dealers. The results in the Appendix show the seeminglyunrelated panel regression results for the absolute value of fails to deliverminus fails to receive in the Treasury and Agency MBS repo markets. Notsurprisingly this difference in the repo market is not driven by demands forliquidity. Instead the market is growing for other structural reasons, e.g., therise of large money managers, and foreign investors (see Gorton and Metrick(2015)).

The Agency MBS breakpoints for the absolute value of the differencebetween fails to receive and fails to deliver are shown in Panel B of Table1. The break points are a bit different from those for Treasuries and AgencyMBS. The regressions for Agency MBS fail to receive and fail to deliver arein the Appendix. For both types of fails the liquidity demand, as captured bythe GC repo-Treasury spread is significant for the second break period, fromSeptember 16, 2009 to December 27, 2011. Throughout the financial crisisthere continued to be a significant amount of agency MBS issued. In 2008and 2009 $2.89 trillion was issued (see Vickery and Wright (2013)). This wasa period of a flight to quality, that is a desire to hold very liquid securities.

2.3 Summary

The evidence for collateral mobility is indirect because there are no data orlimited data on rehypothecation, trading, and collateral posting for deriva-tive positions and for clearing and settlement. Nevertheless, the size of thesecuritization market prior to the crisis and the evidence above, indicate thesystem of mobile collateral that had developed.

3 An Immobile Collateral System

Bolles (1902) described the U.S. National Bank Act as “ . . . the most im-portant measure ever passed by any government on the subject of banking.”The National Bank Acts were passed during the U.S. Civil War; the first Actwas passed in 1863 and this law was amended in 1864. The Act created anew national banking system. The Act was intended to create a demand forU.S. Treasury bonds because without the income tax it was the only way to

13

finance the North in the war. The Acts established a new category of banks,national banks, which were to coexist with state chartered banks. Nationalbanks could issue bank-specific national bank notes by depositing eligibleU.S. Treasury bonds with the U.S. Treasury.9 In this section we examinethe U.S. National Banking Era. In subsection 3.1 we provide a very briefbackground on the banking system in the era, 1863-1914. In subsection 3.2we introduce the “bank note paradox”. We show the “arbitrage profits” thatallegedly existed. The analysis of the profitability of note issuance and its re-lation to the convenience yield on Treasuries is in subsection 3.3. Subsection3.4 summarizes the results of this section.

3.1 The U.S. National Banking System

During the U.S. National Banking Era banks were required to back theirprivately-produced money in the form of bank-specific national bank noteswith U.S. Treasury bonds. One kind of money was required to back anotherkind of money—narrow banking. So, there was a collateral constraint on theissuance of money by banks. As with repo today, the interest on the bondswent to the banks. With national bank notes backed by U.S. Treasuries therewas for the first time in the U.S. a uniform currency. Prior to the Acts, banksissued individual private bank notes which traded at discounts to face valuewhen traded at a distance from the issuing bank. There were hundreds ofdifferent banks’ notes, making transacting difficult. Initially, national banknote issuance was limited to 100 percent of a bank’s paid-in capital, but thiswas changed to 90 percent by the act of March 3, 1865. Also, note issuancewas limited to 90 percent of the lower of par or market value. This waschanged to 100 percent by an act in 1900. See Noyes (1910), Friedman andSchwartz (1963), and Champ (2011c) for more information on the NationalBanking Era.

9Eligible bonds were U.S. Treasury government registered bonds bearing interest incoupons of 5% or more to the amount of at least one-third of the bank’s capital stock andnot less than $30,000. The Act of July 12, 1870 eliminated the requirement that bondsbear interest of 5% or more. After that date eligible bonds were “of any description ofbonds of the U.S. bearing interest in cash.”

14

3.2 The Bank Note Issuance Puzzle

The National Banking Era has been puzzling for economists for well over acentury. The puzzle is that there appears to have been high, allegedly some-times infinite, profits from issuing national bank notes—riskless arbitrageprofits—but this capacity to issue notes was never fully utilized. Friedmanand Schwartz (1963, p. 23): “. . . despite the failure to use fully the possibil-ities of note issue, the published market prices of government bonds bearingthe circulation privilege were apparently always low enough to make noteissue profitable . . . The fraction of the maximum issued fluctuated withthe profitability of issue, but the fraction was throughout lower than mighthave been expected. We have no explanation for this puzzle.” They go onto write: “Either bankers did not recognize a profitable course of action sim-ply because the net return was expressed as a percentage of the wrong base,which is hard to accept, or we have overlooked some costs of bank note issuethat appeared large to them, which seems must more probable” (p. 24).10

Phillip Cagan (1963, 1965) determined whether it was profitable for banksto issue notes by examining the following formula:

r =

{rbp−ταmin(p,1)p−αmin(p,1)

if p > αmin(p, 1)

∞ if p = αmin(p.1)

where: r is the annual rate of return on the issuance of national bank notes; pis the price of the bond held to back the notes (dollars), assuming a par valueof one; rb is the annualized yield to maturity on the bond held as backing; αis the fraction of the value of a given deposit of bonds that could be issued asnotes; and τ is the annual expense in dollars of issuing αmin(p, 1) in notes.The term αmin(p, 1) refers to the amount of notes that are returned to theissuing bank by the U.S. Treasury from the deposit of a bond with price p.The variable τ includes the tax rate on note issuance, which was $0.01 for$1 prior to 1900 and $0.005 on 2% coupon rate bonds after 1900). Also,miscellaneous costs are included here. For example, Cagan used an estimateof these costs of 0.00625 per one-dollar deposit in government bonds.11

10Champ (2011b) also cites these Friedman and Schwartz passages. Friedman andSchwartz’s mention of “the wrong base” refers to mistaken calculations by the contempo-rary Comptroller of the Currency.

11The Comptroller used $62.50 for the costs associated with notes issued based on$100,000 of bonds deposited. These costs included the cost of redemption, $45; expresscharges, $3; engraving plates for the notes, $7.50; and agents’ fees, $7. See Champ

15

Champ (2011b) gives the following example. Consider a bank in 1890(i.e, α=0.9) that purchased a bond for $1.10, with yield to maturity of 4percent. Then, the total cost of note issuance is τ = 0.01 + 0.00625

0.09≈ 0.01694.

So, in this case. the rate of profit for issuing notes backed by this bond soldbe:

r ≈ (0.04)(1.10)− (0.01694)(0.9)

1.10− 0.9≈ 14.375%

Cagan (1963) found very high profits rates for the 1870s, 20-30%. Moreimportantly, Cagan and Goodhart (1965) found profit rates of infinity in theearly 1900s. An infinite rate of profit occurs when α = 1 after 1900 and thebond is selling below par. In that case, the notes the bank could issue basedon using that bond as collateral would exactly equal the price paid for thebond, so no capital could be used and the bank could earn infinite profits.Figure 7 shows Cagan’s profit series (as computed by Champ (2011b)). Thefigure shows the conundrum. The gaps in the figure are the instances wherethe rate of profit was alleged to be infinity; these are all cases where theprices of 2% coupon bonds fell below par. Note that the figure calculates theaverage rate of profit, not the marginal rate of profit. National banks didnot take advantage of apparent profit opportunities.

Figure 8 shows our calculation of the profit series. We used Champ’s moreaccurate representation of the costs of note issuance than the Comptroller ofthe Currency.12 We also filled in all the bond prices that were missing froma Bruce Champ spreadsheet (provided to us by the Federal Reserve Bank ofCleveland). We also made further adjustments discussed below.

Why didn’t banks take advantage of this arbitrage opportunity? Afterall, banks held U.S. Treasury bonds on their balance sheets (i.e., not includ-ing Treasuries held to back their notes). See Figure 6. Or, banks could haveraised capital and used this to buy bonds for collateral for notes. To explainthis puzzle, the literature has focused on hidden transactions costs or therisks of unpredictable redemptions; see Bell (1912), Cagan (1965), Good-hart (1965), Cagan and Schwartz (1991), Duggar and Rost (1969), Champ,Wallace and Weber (1992), and Wallace and Zhu (2004). None of these ex-planations are particularly persuasive; these explanations are reviewed by

(2011b).12Based on a spreadsheet of Bruce Champ, provided by the Federal Reserve Bank if

Cleveland.

16

Champ (2011b) and Calomiris and Mason (2008).13

Prior explanations do not mention that there may be a convenience yieldassociated with U.S. Treasuries. This is, however, suggested by the workof Krishnamurthy and Vissing-Jorgensen (2012) who look at data over theperiod 1926-2008, and Krishnamurthy and Vissing-Jorgensen (2013) who an-alyze the period 1914-2011. Figure 6, showing that national banks held U.S.Treasuries on their balance sheets (i.e., the ratio of U.S. government bondsnot on deposit at the Treasury to bank loans and discounts) suggests, byrevealed preference, that there was a convenience yield associated with Trea-suries during the National Banking Era.

At the time bankers also recognized the convenience yield on Treasuries.For example, The Financier, April 7, 1902, Volume LXXIX (The FinancierCompany): “. . . banks have always regarded high class bonds as anoffset, so to speak, for risks incurred in discounts yielding a higher rate ofinterest. In this connection we cannot do better than to quote from a veryvaluable paper read by A.M. Peabody, of St. Paul, before the St. Paul BankClerks’ Association, in which this feature is brought prominently forward.After explaining the classification of such investments, Mr. Peabody says:’They have ever proved themselves the safeguards for banks under pressure offinancial panics in times of great stringency, and when it would be impossibleto borrow money on any form of security, railroad bonds with governmentbonds, are alone available as security for money’ (p. 1258). And, The BondBuyers’ Dictionary (1907): “. . . it is possible to say that there is a bettermarket in moments of extreme panic for the Government issues than thereis for even the best class railroad bonds. There will not be by any means[be] the same volume of liquidation. For every dollar of Government bondsthrown into a panic market there will be $100 of railroad bonds. . . .Government bonds are undoubtedly the safest of all securities. . . “ (p. 73).

But, even if banks wanted to keep these Treasuries on their balance sheets,why didn’t they raise bank capital to buy Treasuries to back note issuance.That they did not suggests that bank capital was costly or that banks couldnot find the bonds. We will argue that bankers did not take advantage ofopportunity to issue more national bank notes because it was not profitableto do so. We will show that the implicit profit from not issuing notes is

13Calomiris and Mason (2008) show that there was “. . . substantial variation inthe propensity to issue national bank notes traceable to county, state, and bank-specificcharacteristics related to the profitability of lending” (p. 340).

17

driven by measures of convenience yield.We first return to the calculation of the profit rate from note issuance.

As mentioned above, we filled in the missing bonds in Champ’s originalspreadsheet used for calculating the profit rate to note issuance.14 We nexteliminated bonds that would have been called in the next six months, sincethen the notes backed by these bonds will have to be returned, or new bondswill have to be purchased.15

However, there is another issue, namely that it is the marginal profit ratesthat are relevant not the average rate of profit. This is important because inthe early 1900s, and possibly before that. U.S. government bonds were hardto find. And, even when banks could find bonds, they had to reverse repoin the bonds in at a high cost. Contemporary observers continually wroteabout this shortage of safe debt. For example, Morris (1912): “Variousreasons have been assigned for the decline in circulation which culminated in1891, the most probable being the growing scarcity of U.S. bonds and theirrelatively high premium. It is also alleged that improved banking facilities,allowing a more extensive use of checks, reduced the demand for currency” (p.492). Morris dates the start of the problem as 1891. It is also interesting thatMorris points out that the cost of note issuance caused a further developmentof demand deposits, the shadow banking system of its time.

Borrowing bonds was costly. Francis B. Sears, vice president of the Na-tional Shawmut Bank of Boston, Mass. (1907-08): “There are two classes ofbanks—those outside of the large cities, that can get bonds only by buyingthem, and a few banks in a few large cities that can borrow them. I wouldlike to add that insurance companies and savings banks are large bondhold-ers, and undoubtedly arrangements can be made with them to get bondsfor some large banks. The rate is 11

2to 2 percent for borrowing bonds in

that way” (p. 91). Bankers Magazine (March 1908): “Bond borrowings bythe national banks have become an important feature of banking in recentyears. Where a bank wishes to increase its circulation, or to procure publicdeposits, and does not happen to have the bonds which must be pledged withthe Treasury, and finding the market price of bonds too high to make the

14The missing bond prices/amounts were mostly during 1875-1879 plus one bond ma-turing in 1896. This did not affect the potentially infinite profits, but just added moreobservations in the earlier period.

15Eliminating these bonds removed some spikes in the profit series, one of which wasduring the period of high profits (1907). But otherwise it has no significant effect on thepost-1902 series.

18

transaction profitable if the bonds must be bought, resort is had to borrow-ing. Bond dealers, savings banks or private holders may have ‘Governments’which they are willing to lend to national banks for a consideration” (p.321).16 The Rand-McNally Bankers’ Monthly (September, 1902) quoting“a banker”: “There is not much profit in issuing circulation on governmentbonds, but some of the larger banks are willing to take out notes, if theycan borrow bonds for that purpose from their friends—not being disposedto buy them for temporary use. . . . The real trouble is to find the bonds.Many of them are held by institutions and estates, who cannot legally loanthe bonds to National Banks, and as their prices are too high to justify anylarge purchases of bonds by banks for the purpose of taking out circulation.. .”(p. 157-158). Gannon (1908), speaking of Treasury bonds: “. . . suchbonds are not easy to buy in quantity, and the greater part of the recentexpansion, some $80,000,000 since the panic [of 1907], was accomplished byborrowing bonds”(p. 338)

The situation was summarized by The Financial Encyclopedia (1911, p.119):

When the banks borrow, either to secure banknote circulationor Government deposits, they make private arrangements withthe actual owners of the bonds, including insurance companies,for the use of these securities. The rates banks pay vary, but ingeneral lenders of bonds secure a very substantial profit from thisemployment of them, in addition to the interest which the bondsthemselves carry.

Borrowed bonds were first itemized separately in the nationalbanks’ returns under the Comptroller’s call of November 25, 1902.At that time the total ‘borrowed bonds’ reported by nationalbanks of the whole country were $39,254,256 of which New Yorkbanks were credited with $21,199,000. In the return of Decem-ber 3, 1907, the banks of the United States reported bonds of$166,073,021, more than half, or $88,274,330, being held by theforty national banks of this [New York] city. These are by far thelargest holdings ever reported by New York banks.

When a bank borrows Government, municipal, or other bonds,

16Government deposits in national banks had to be backed by bonds also, but there wasa slightly larger list of eligible bonds for this purpose.

19

from an insurance company, for instance, which are pledged assecurity for public (Treasury) deposits, it either gives the lendera check for the face value, with a contract stipulating to buy backthe bonds at a certain price, or the bank gives the lender othercollateral as security for the loan.

In the case of life insurance companies, the collateral offered inexchange for the bonds has often represented bonds in which thelending corporations are allowed to invest, but which were notin the so-called ‘savings bank list,’ and for that reason were noteligible as security for public deposits. While one or two of thelife companies have never consented to lend their bonds, manyothers, as well as various fire insurance companies, have done so,on the theory that it was a good business transaction, since ityielded them 1 or 11

2% in addition to the regular interest return.

The scarcity of bonds meant that the marginal cost of conducting the“arbitrage” was higher than the average cost. While the Comptroller startedpublishing data on bank bond borrowings in 1902, it seems that this problemstarted earlier. At a meeting of the American Economic Association held inCleveland, Ohio in December 1897, it was voted to appoint a committeeof five economists to consider and report on currency reform in the UnitedStates.17 They turned in a report in December 1898. One point they madewas this:“Now it is commonplace that our bank circulation is not a veryprofitable one.” See The Bankers’ Magazine, February 1899, p.221.

Note issuance profit series that are the average rate of profit are mislead-ing. To adjust the profit calculations to reflect the scarcity and associatedhigh cost of reversing in bonds, we set α in the above calculation of the profitrate to 0.99 instead of 1. Now, there are no instances of infinite profits. Wediscuss below why this does not greatly affect regression results. Figure 8shows the series of profit rates in this case.

There is also the issue of the cost of bank capital. This cost is hard toquantify, as it is today. Bank stock during this period was illiquid, tradingon the curb market. And there is some evidence that it was held in blocks byinsiders. See Gorton (2013). There is no data on bank stock issuance. Con-temporaries described the return to bank stock as low, partly due to double

17The economists were a very distinguished group: F.M. Tayor, University of Michigan;F.W. Taussig, Harvard; J.W. Jenks, Cornell; Sidney Sherwood, Johns Hopkins; and DavidKinley, University of Illinois.

20

liability.18 For example, Frank Mortimer, cashier of the First National Bank,Berkeley. Ca.; in an address delivered before the San Francisco Chapter ofthe American Institute of Banking, American Institute of Banking Bulletin“When one takes into consideration the risk involved, the capital invested,and the double liability attached to stockholders in national banks, the profitfrom an investment in bank stock is small, indeed, when compared to theprofit accruing from other lines of business.” (p. 236; reprinted in the Journalof the American Bankers Association, vol. 6, July 1913-June 1914.).

3.3 The Convenience Yield on Treasuries and the Costof Bank Capital

In this section we turn to an analysis of the rate of profit on note issuance.We show that the rate of profit on note issuance is highly related to theconvenience yield on Treasuries. We measure convenience yield in two com-plementary ways. First, we use the supply of Treasuries divided by GDP. Kr-ishnamurthy and Vissing-Jorgensen (2012) show that this measure stronglydrives the convenience yield on Treasuries from 1926-present. When thesupply of those assets is low, that is safe assets are relatively scarce, thenthe convenience yield for safe assets increases. Therefore, Treasury supplyshould be negatively related to the convenience yield. We take two measuresof Treasury supply: (US government debt)/GDP (as in Krishnamurthy andVissing-Jorgensen (2012)), and (available Treasuries)/GDP, where availableTreasuries excludes those already held to back bank note issuance and thuscaptures the remaining supply. Second, we also measure the convenienceyield as the spread between high grade municipal bonds from New Eng-land and Treasuries. Municipal bond yields are from Banking and MonetaryStatistics (1976).

Table 4 gives the results of a regression of issuance profits on these mea-sures of the convenience yield from 1880-1913. The results match our in-tuition. The profit measure is high exactly when the convenience yield toTreasuries is large. We find that a 1% increase in the muni spread is as-sociated with a 15% increase in average profit. As demand for Treasuriesincreases, the apparent profits also increase. As the supply of available Trea-suries decreases, profits also increase. Both the supply variables and the munispread are highly significant independently. However, we would suspect that

18On double liability see Macey and Miller (1992) and Grossman (2001).

21

they likely measure similar economic forces though each is measured withnoise. Consistent with this, when we include both the supply of Treasuriesand muni spread together the coefficients on each decrease in absolute valuethough they remain statistically significant. This suggests that both areimperfect but overlapping measures of the convenience yield.

We show the results when we use the average profit series as well as the logaverage profit series. Recall that profit is given by rbp−ταmin(p,1)

p−αmin(p,1). A possible

concern is that the profit series is highly non-linear due to the denominatorbecoming small later in the sample. To mitigate this concern, we also reportthe results using log profits, which largely alleviates the strong non-linearityin the denominator (see Figure 8 which plots profits on a log scale). Ourresults do not change drastically with the log transformation, highlightingthat non-linearities in the latter half of the sample aren’t driving the result.In unreported results we also obtain the same basic findings for alternativevalues of α.

All variables in this regression are persistent which can potentially con-found inference. We deal with this in several ways. First, in our mainspecifications we estimate standard errors using Newey-West with 10 yearlags (specifically, we use 10 lags for annual data and 40 lags for quarterlydata). Second, we run GLS assuming the error term follows an AR(1). Thissuggests transforming both our x and y variables by 1 − ρL where ρ is theerror auto-correlation and L is a lag operator. We find ρ by running OLS asin specification (4) in the Table and computing the sample auto-correlationof the residuals. This does not substantially change the point estimates orinference in terms of what is statically significant. As mentioned by Krish-namurthy and Vissing-Jorgensen (2012), however, coefficients do decreasesomewhat in absolute value. A likely reason is that these are noisy measuresof convenience yield and measurement error will become more pronounced inthe transformed data. This follows from the fact that when x is persistentthe variance of x will be dominated by low frequency components. In con-trast, in the transformed data measurement error likely accounts for more ofthe variance of the right hand side variable, resulting in a larger degree ofattenuation bias.

Finally, while the results appear fairly strong, we also acknowledge thatwe are working with a fairly small subsample of data which is a limitation ofour analysis. Higher frequency data (e.g., monthly data on debt/GDP) won’tbe particularly helpful here in overcoming the fairly small sample because thevariables are highly persistent.

22

We hypothesize that a third variable – the cost of capital for banks –likely plays a role in explaining the profits on note issuance as well. If thecost of raising capital for banks is high, then banks would find it costly totake advance of note issuance and may leave a puzzlingly large profit on thetable. For this conjecture, we can only offer suggestive evidence from Figure8 which plots the profit series along with NBER recession bars. It is likelythat the cost of raising capital for banks increases during recessions, andespecially at the onset of recessions, and these are times when we do in factsee increases in the profit series. Thus, there is some suggestive evidence ofthe cost of capital for banks being positively associated with the profits onnote issuance as well.

Taken together, our results indicate that the profits to note issuance fluc-tuate with the convenience yield on Treasuries, and our evidence is consistentwith the idea that profits are related to the cost of bank capital.

3.4 Proposals for Reform during the National BankingEra

Despite the creation of a uniform currency, the monetary system of the Na-tional Banking Era was increasingly unpopular. Reform of the currencysystem was increasingly discussed because of the problems of inelasticity andbanking panics. “Inelasticity” meant that the quantity of money was notsensitive to the business or seasonal cycles. There were spikes in seasonalinterest rates and the money supply could not be increased to alleviate bankruns. Following the Panic of 1893 calls for reform of the banking systembecame louder. All the reforms sought to sever the link between bank notesand Treasuries. This link was viewed as the problem. Replacement proposalswere for an “asset-based” system, meaning that currency would be allowedto be backed by bank loans or commercial paper (depending on the partyproposing this system).

The first proposal for an asset-backed currency came from the AmericanBankers’ Association in 1894. It was called the “Baltimore Plan” (the As-sociation’s meeting was in Baltimore that year). The plan envisioned banksissuing circulating notes under federal supervision, where the notes wouldbe secured by (1) a first lien upon the assets of the issuing bank; (2) thedouble liability of shareholders; (3) the 5 percent redemption fund; and (4) a5 percent guaranty fund (to be used to repay note holders of a failed bank).

23

Treasury Secretary Carlisle (1894) modified the Baltimore Plan to createmore security for the noteholders. He proposed (1) a deposit of 30 percent ofthe circulation in legal money with the government; (2), the safety fund of 5percent; (3) a requirement upon all other banks to cover any losses beyondthe two funds of any failed bank; (4) a lien upon all assets; and (5) a furtherlien on the shareholders’ liability. In addition, both plans wanted to limitnote issuance to 50 or 75 percent of paid-up capital.

There were other subsequent proposals as well, including plans from Eck-els (1894a,b) and the Indianapolis Monetary Commission (1900), see Taylor(1898). All of these plans were similar in spirit to the Baltimore Plan. SeeWicker (2005), West (1974) and Laughlin (1920) for more detailed discussionsof the various plans.

The asset-based monetary proposals sought to separate the two forms ofmoney: Treasuries and bank notes. And they did recognize that bank runson demand deposits were related to perceptions of the “safeness” of the loansbacking the demand deposits. Backing private bank notes with loan portfo-lios would have created the risk of runs on notes as well as demand deposits,although seniority of notes, guarantee funds, reserves, etc. were aimed at thisproblem. But, the proposals embedded the view that an “elastic” currencywould alleviate panics. Laughlin (1920): ”In all these plans we were reallyaiming to prevent the difficulties experienced in [the Panic of] 1893. . . ” (p.30-31).

It is interesting that the proposed reforms during the period all aimed tobreak the link between bank notes and Treasuries.

4 Discussion

Of course, the National Banking System is not exactly like the LCR. Thereare obvious differences. But, like the National Banking Era, the logic ofthe LCR seems to be that if short-term debt is backed by Treasuries, thenbank runs will be avoided. Fundamentally the two systems enforce a corre-spondence between two types of debt instruments, each with a convenienceyield. The input for making one kind of money, bank notes or money marketinstruments, is Treasuries. Such a system is fragile because by forcing twokinds of money together it is likely that there will be a shortage of one kindof money, leading to its private production elsewhere, which creates fragilityin the system.

24

The National Banking System did succeed in introducing a uniform cur-rency where the national bank notes of different banks all traded at par,unlike the pre-Civil War period. This was because banks’ national banknotes were backed by U.S. Treasury debt. We have focused on the fact thatTreasuries have a convenience yield. In the recent period, a measure of thatconvenience yield is the GC repo to Treasury spread. This spread is a driverof repo fails when agents want to keep the Treasuries (or the cash) in repotransactions.

This was the same core issue during the National Banking Era. In theNational Banking Era, a measure of the convenience yield (and the cost ofbank capital) is the implied profitability of issuing bank notes. There is nounder issuance puzzle once this is recognized. If Treasuries have a convenienceyield, then they provide safety and liquidity to the agents who demand this.But, these agents or other agents also have a demand for bank notes.

If there were enough Treasuries (high-quality liquid assets) to meet theglobal demand for safe debt and to back short-term bank debt, then theLCR and related immobile collateral requirements would not be a problem.One potential argument is that in the National Banking Era the supply ofTreasuries was low (debt to GDP was in the range of 10-30%) so that scarcitywas more of an issue in that period then it is today where the supply ofgovernment debt is much larger. However, this ignores that the demandfor safe US government debt now is also global, which can add to issuesof scarcity. The likelihood of such a satiation of the global economy withTreasuries today seems remote. Gorton, Lewellen and Metrick (2012) showthat the sum of U.S. government debt outstanding and privately-producedsafe debt outstanding has been 32 percent of total assets in the U.S. since1952. See Figure 10. The figure shows that the bulk of safe debt has neverbeen Treasuries, but has mostly been privately produced debt. (Figure 1shows the composition of this privately produced safe debt.) There has neverremotely been enough U.S. Treasuries to make up the 32 percent and giventhe debt burden of issuing enough to accomplish that, there is never likely tobe enough. Furthermore, Treasuries outstanding is a function of fiscal policynot a function of the demands for collateral.

When Treasuries have a convenience yield, and short-term bank debtmust be backed by Treasuries, there is a tradeoff between the two types ofmoney. More short-term debt means fewer Treasuries for alternative uses.This tradeoff is common to the two systems. We saw this in the NationalBanking Era. The tradeoff is evident in the data. Noyes (1910): “A heavy

25

decrease in the outstanding public debt would naturally, at some point, causea reduction in the bank-note circulation, independently of other influences. Alarge increase in the government debt would necessarily cause an increase inthe supply of bank notes” (p. 4). Noyes then traces this out over the NationalBanking Era. It is the same statement that was formalized by Krishnamurthyand Vissing-Jorgensen (2012a,b) for the modern era. The convenience yieldis negatively related to Treasuries (divided by GDP) outstanding.

One way out of this tradeoff, if it is binding, is to privately produceanother kind of debt. In the recent period this was ABS and MBS, whichcould be used in place of Treasuries to back repo, ABCP, and MMF. In theNational Banking Era, it was demand deposits using portfolios of loans as thebacking. During the National Banking Era, Treasuries outstanding to GDPfell secularly (see Figure 6, panel C). And, from the start of the NationalBanking System, the ratio of bank notes to demand deposits fell, from justover 60 percent in the early part of 1865 to 14 percent by 1909, as shown inFigure 9. Demand deposits were privately-produced safe debt or money, theshadow banking system of its time. So, while the immobile collateral systemended bank runs on bank notes, there were bank runs on demand deposits,another form of bank money. The biggest problem of the National BankingEra was that there were banking panics.

It is difficult to prove causally that demand deposits grew relative to banknotes because of collateral requirements. However, the growth in depositsdoes line up remarkably well with the decline in Treasury supply and hencethe supply of safe assets that could be used to back notes. This accords withthe evidence in Krishnamurthy Vissing-Jorgensen (2012b) that the supplyof Treasury crowds out privately produced bank debt. The ratio of nationalbank notes to demand deposits fell from over 60% to less than 20% over theperiod and this ratio co-moves strikingly with the debt to GDP ratio as shownin Figure 9. The correlation of the ratio of notes to deposits with the supplyof Treasuries to GDP is 0.96. As the supply of Treasuries falls over thisperiod, deposits grow. Demand deposits were conceptually misunderstoodduring the National Banking Era, although it was clear that demand depositswere the issue in banking panics. This highlights that the system of immobilecollateral in the National Banking Era was not successful in mitigating panicsand that it likely contributed to the growth of other forms of bank debt.

26

References

Bai, Jushan (2010), “Common Breaks in Means and Variances for PanelData,” Journal of Econometrics 157, 78-92.

Baltimore Plan (1894), American Banking Association, Proceedings.

Bank for International Settlements (BIS), Bank Commitee on Banking Su-pervision (2013), ”Basel III: The Liquidity Coverage Ratio and LiquidityRisk Monitoring Tools.”

Basel Committee on Banking Supervision, Bank for International Settle-ments (2011), “Basel III Framework for Liquidity–Frequently Asked Ques-tions.”

Bell, Spurgeon (1912), “Profit on National Bank Notes,” American EconomicReview 2, p. 38-60.

Bernanke, Ben S. (2007), “Global Imbalances: Recent Developments andProspects,” speech delivered at the Bundesbank Lecture, Berlin, Germany,September 11.

Bernanke, Ben S. (2005), “The Global Savings Glut and the U.S. CurrentAccount Deficit,” speech delivered at the Sandridge Lecture, Virginia Asso-ciation of Economists, Richmond, Va., March 10.

Bernanke, Ben S., Carol Bertaut, Laurie Pounder DeMarco and Steve Kamin(2011), “International Capital Flows and the Returns to Safe Assets in theUnited States, 2003-2007,” Working paper, Board of Governors of the FederalReserve Sysftem.

Bertaut, Carol, L.P. DeMarco, S. Kamin and R. Tryon (2012), “ABS Inflowsto the United States ant the Global Financial Crisis,” Journal of Interna-tional Economics 88, 219-234.

Bertaut, Carol, Alexandra Tabova and Vivian Wong (2014), “The Replace-ment of Safe Assets: Evidence from the U.S. Bond Portfolio,” Board of Gov-ernors of the Federal Reserve System, working paper.

Board of Governors of the Federal Reserve System (1976), Banking and Mon-etary Statistics

27

Bolles, Albert (1902), “Responsibility of the National Bank in the PresentCrisis,” Annals of the American Academy of Political and Social Science,Vol. 20, 1-18.

Bond Buyers’ Dictionary, The (1907), edited by S.A. Nelson, (S.A. Nelson& Co.; New York).

Caballero, Ricardo (2010), “The ’Other’ Imbalance and the Financial Crisis,”NBER Working Paper 15636.

Caballero, Ricardo and Arvind Krishnamurthy (2009), “Global Imbalancesand Financial Fragility,” American Economic Review: papers & Proceedings99, 584-588.

Cagan, Phillip (1963), “The First Fifty Years of the National Banking Sys-tem,” in Banking and Monetary Studies, edited by Deane Carson, 15-42(Richard D. Irwin; Homewood, IL).

Cagan, Phillip (1965), Determinants and Effects of Changes in the Stock ofMoney, 1875-1960 (NBER; New York).

Cagan, Phillip and Anna J. Schwartz (1991), “The National Bank Note Puz-zle Reinterpreted,” Journal of Money, Credit and Banking 23, 293-307

Calomiris, Charles and Joseph Mason (2008), “Resolving the Puzzle of theUnderissuance of National Bank Notes,” Explorations in Economic History45, 327-355.

Carlisle, John (1894), Report of the Committee on Banking and CurrencyTogether with Hearings, House of Representatives, 8149, 53rd Congress, 3rdSession.

Champ, Bruce (2011a), “The National Banking System: The Empirical Ev-idence,” Federal Reserve Bank of Cleveland, Working Paper 07-19R.

Champ, Bruce (2011b), “The National Banking System: The National BankNote Puzzle,” Federal Reserve Bank of Cleveland, Working Paper 07-22R.

Champ, Bruce (2011c), ”The National Banking System: A Brief History,”Federal Reserve Bank of Cleveland, Working Paper 07-23R.

28

Champ, Bruce, Neil Wallace, and Warren Weber (1992), “Resolving the Na-tional Bank Note Paradox,” Federal Reserve Bank of Minneapolis QuarterlyReview (Spring), 13-21.

Duffie, Darrell, Martin Scheicher and Guillaume Vuillemey (2014), “CentralClearing and Collateral Demand,” Journal of Financial Economics, forth-coming.

Duggar, Jan Warren and Ronald Rost (1969), “National Bank Note Redemp-tion and Treasury Cash,” Journal of Economic History 29, 512-520.

Dunbar, Charles (1887), “Deposits as Currency,” Quarterly Journal of Eco-nomics 1, 401-419.

Eckels, James (1894a), Report of the Committee on Banking and CurrencyTogether with Hearings, House of Representatives 8149, 53rd Congress, 3rdSession.

Eckels, James (1894b), Annual Report of the Comptroller of the Currency,3rd Session, 53rd Congress, Dec. 3, Government Printing Office, WashingtonDC.

Fisher, Willard (1895), “Money and Credit paper in the Modern Market,”Journal of Political Economy 3, 391-413.

Garbade, Kenneth and Michael Fleming (2005), “Explaining SettlementFails,” Federal Reserve Bank of New York Current Issues in Economics andFinance, vol. 11, no. 9 (September).

Fleming, Michael and Kenneth Garbade (2002), “When the Back OfficeMoved to the Front Burner: Settlement Fails in the Treasury Market after9/11,” FRBNY Economic Policy Review (November), 35-57.

Fender, Ingo and Ulf Lewrick (2013), “Mind the Gap? Sources and Im-plications of Supply-Demand Imbalances in Collateral Asset Markets,” BISQuarterly Review, September 2013, 67-81.

Friedman, Milton and Anna Scwartz (1963), A Monetary History of theUnited States (Princeton University Press).

29

Garbade, Kenneth, Frank Keaane, Lorie Logan, Amanda Stokes and JenniferWolgemuth (2010), “The Introduction of the TMPG Fails Charge for U.S.Treasury Securities,” FRBNY Economic Policy Review (October), 45-71.

Goodhart, Charles A. E. (1965), “Profit on National Bank Notes, 1900-1913,”Journal of Political Economy 73, 516-522.

Gorton, Gary (2013), ”The Development of Opacity in U.S. Banking,” YaleJournal of Regulation, forthcoming.

Gorton, Gary and Andrew Metrick (2015), ”Who Ran on Repo,” Yale work-ing paper.

Gorton, Gary, Stefan Lewellen, and Andrew Metrick (2012), “The Safe-AssetShare,” American Economic REview: Papers & Proceedings 102, 101-106.

Gourinchas, Pierre-Olivier and Olivier Jeanne (2012), “Global Safe Assets,”paper prepared for the XI BIS Annual Conference held in Lucerne, June20-21, 2012.

Grossman, Richard (2001), ”Double Liabioity and Bank Risk Taking,” Jour-nal of Money, Credit and Banking 33(2), 143-159.

Hammond, Bray (1957), Banks and Politics in America from the Revolutionto the Civil War (Princeton University Press).

Heller, Daniel and Nicholas Vause (2012), “Collateral Requirements forMandatory Central Clearing of Over-the-Counter Derivatives,” BIS Work-ing Papers No. 373.

Indianapolis Monetary Commission (1920), Report of the Monetary Commis-sion of the Indianapolis Convention (Hollenbeck Press; Indianapolis).

James, John A. (1976), “The Conundrum of the Low Issue of National BankNotes,” Journal of Political Economy 84, 359-368.

Kinley, David (1910), The Use of Credit Instruments in Payments in theUnited States (National Monetary Commission; Washington D.C.).

Krishnamurthy, Arvind and Annette Vissing-Jorgensen (2012a), “The Aggre-gate Demand for Treasury Debt,” Journal of Political Economy 120, 233-267.

30

Krishnamurthy, Arvind and Annette Vissing-Jorgensen (2012), “Short-termDebt and Financial Crisis: What can we learn from U.S. Treasury Supply,”working paper.

Kuehlwein, Michael (1992), “The National Bank Note Controversy Reexam-ined,” Journal of Money Credit and Banking 24, 111-126.

Laughlin, Lawrence (1920), Banking Progress (Charles Scribners; New York).

Lucas, Robert (1976), “Econometric Policy Evaluation: A Critique,” inK. Brunner and A. Meltzer, eds., The Philips Curve and Labor Markets,Carnegie-Rochester Conference Series on Public Policy 1 (Elsevier; NewYork), 19-46.

Macaulay, Frederick (1958), The Movements of Interest Rates, Bond Yieldsand Stock Prices in the United States Since 1856 (National Bureau of Eco-nomic Research; New York, NY).

Macey, Jonathan and Geoffrey Miller (1992), ”Double Liability of BankShareholders: History and Implications,” Wake Forest Law Review 27, 31-62.

Maggiori, Matteo (2013), “Financial Intermediation, International Risk Shar-ing, and Reserve Currencies,” Harvard University, working paper.

Noyes, Alexander (1910), A History of the National Bank Currency, NationalMonetary Commission (Government Printing Office; Washington D.C.

Sidanius, Che and Filip Zikes (2012), “OTC Derivatives Reform and Col-lateral Demand Impact,” Bank of England, Financial Stability Paper No.18.

Singh, Manmohan (2011), “Velocity of Pledged Collateral,” IMF WorkingPaper WP/11/256.

Singh, Manmohan and James Aitken (2010), “The (sizeable) Role ofRehypothecation in the Shadow Banking System,” IMF Working PaperWP/10/172.

Taylor, F. M. (1898), “The Final Report of the Indianapolis Monetary Com-mission,” Journal of Political Economy 6, 293-322.

31

Treasury Market Practices Group (2011), “Understanding Settlement Failsin Agency Mortgage-Backed Securities.”

Vickery, James and Joshua Wright (2013), “TBA Trading and Liquidity inthe Agency MBS Market,” Federal Reserve Bank of New York EconomicPolicy Review (May), 1-18.

Wallace, Neil and Tao Zhu (2007), “Float on a Note,” Journal of MonetaryEconomics 54, 229-246.

West, Robert Craig (1974), Banking Reform and the Federal Reserve, 1863-1923 (Cornell University Press; Ithaca, NY and London).

Wicker, Elmus (2005), The Great Debate in Banking Reform: Nelson Aldrichand the Origins of the Fed (Ohio State University Press; Columbus, OH).

Wicker, Elmus (1966), Federal Reserve Monetary Policy 1917-1933 (RandomHouse:New York).

5 Tables and Figures

32

Table 1: We present summary statistics on fails and rehypothecation.Sources: Panels A & B, Federal Reserve Bank of New York. Panel C, ISDA.

Panel A: Fails, $ Millions

Fail to Receive Fail to Deliver1990-99 2000-07 1990-99 2000-07

Mean 31,676 154,600 11,812 122,363Std Dev 3,771 6,372 4,105 163,564

Panel B: Receive fails minus deliver fails, $ Millions

Treasuries Agencies MBS1990-99 2000-07 1990-99 2000-07 1990-99 2000-07

Mean 5,169 16,624 -123 338 207 -8,073Std Dev 3,947 18,268 739 3,696 2,615 22,283

Panel C: Amount of collateral received eligible vs actually rehypothecated (12/31/2013)

Treasuries OtherTotal Received ($ Millions) 179,366 123,915

Eligible for Rehypothecation 85% 55%Actually Rehypothecated 55% 30%

33

Table 2: We report breakpoints for our fails data panel for both failures toreceive and deliver along with 95% confidence intervals. The methodologyfor finding breaks in panels follows Bai (2010).

Panel A: Treasury and Agency Bonds

Break Date Lower Bound Upper BoundFirst Break 12-Sep-01 2-May-01 16-Jan-02Second Break 24-Sep-08 11-Jun-08 31-Dec-08Third Break 11-Feb-09 14-Jan-09 4-Mar-09

Panel B: Agency MBS

Break Date Lower Bound Upper BoundFirst Break 16-Oct-02 18-Sep-02 6-Nov-02Second Break 16-Sep-09 26-Aug-09 30-Sep-09Third Break 28-Dec-11 23-Nov-11 25-Jan-12

34

Table 3: We run seemingly unrelated regressions of Treasury and Agencyfails to receive.

∆ Fails Rec ∆ Fails Rec ∆ Fails Rec ∆ Fails Rec ∆ Fails Rec ∆ Fails RecGC Repo-1m T-bill 6.963∗∗∗ 0.695 7.303∗∗∗ 0.640 7.509∗∗∗ 0.620

(5.57) (0.41) (5.78) (0.38) (5.91) (0.36)

L1.GC Repo-1m T-bill 2.609∗ 0.818 2.951∗ 0.648(2.07) (0.48) (2.31) (0.38)

L2.GC Repo-1m T-bill 2.495∗ 0.316(1.96) (0.19)

GC Repo-1m T-bill x Break 1 13.35∗∗∗ 13.96∗∗∗ 13.35∗∗∗

(5.14) (5.26) (5.03)

L1.GC Repo-1m T-bill x Break 1 2.492 1.894(0.95) (0.71)

L2.GC Repo-1m T-bill x Break 1 -2.164(-0.82)

GC Repo-1m T-bill x Break 2 39.57∗∗∗ 45.66∗∗∗ 44.08∗∗∗

(7.36) (8.46) (8.20)

L1.GC Repo-1m T-bill x Break 2 33.27∗∗∗ 37.98∗∗∗

(6.55) (7.46)

L2.GC Repo-1m T-bill x Break 2 32.26∗∗∗

(6.43)

GC Repo-1m T-bill x Break 3 -1.485 -1.878 -1.185(-0.13) (-0.16) (-0.10)

L1.GC Repo-1m T-bill x Break 3 4.103 4.818(0.36) (0.41)

L2.GC Repo-1m T-bill x Break 3 3.120(0.27)

Break 1 (9/2001-9/2008) -8.852 -8.269 -6.430(-0.23) (-0.21) (-0.17)

Break 2 (9/2008-2/2009) 75.43 153.1 193.5(0.57) (1.16) (1.48)

Break 3 (2/2009) 11.00 9.994 7.266(0.26) (0.24) (0.17)

D. 1m T-Bill -13.08∗∗∗ -11.39∗∗∗ -9.288∗∗∗

(-6.41) (-5.53) (-4.48)

Constant 0.879 -11.81 1.104 -10.05 1.092 -7.817(0.05) (-0.48) (0.07) (-0.41) (0.06) (-0.32)

Observations 2398 2398 2386 2386 2374 2374R2 0.013 0.055 0.015 0.076 0.016 0.095

t statistics in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

35

Table 4: We run regressions of profits on two measures of the convenienceyield: Treasury supply and the Municipal bond - Treasury spread. We mea-sure Treasury supply as either debt / GDP or Treasuries available / GDP.Treasury supply variables are annual which reduces the number of observa-tions. The muni spread is quarterly. T-stats are Newey-West with 10 lags forannual regressions and 40 lags for quarterly regressions. The column “GLS”assumes errors follow an AR(1) and hence transforms the x and y variablesby 1− ρL where ρ is the autocorrelation of the error term in (4) using OLSand L is a lag operator.

Profit on Convenience Yield, 1880-1913yt = a+ b× xt + εt

Panel A: y =profit

GLS(1) (2) (3) (4) (5)

ln(Debt/GDP) -43.4[-2.05]

ln(Avail/GDP) -31.8 -24.4 -19.3[-3.51] [-4.03] [-4.33]

Muni spread 54.5 14.9 18.3[3.06] [3.59] [3.20]

AdjR2 0.28 0.65 0.50 0.76 0.61N 34 34 137 34 34

Panel B: y = ln(profit)

GLS(1) (2) (3) (4) (5)

ln(Debt/GDP) -1.78[-2.35]

ln(Avail/GDP) -1.18 -1.03 -0.81[-4.85] [-6.81] [-5.49]

Muni spread 1.81 0.44 0.38[3.83] [4.49] [2.05]

AdjR2 0.36 0.67 0.48 0.74 0.54N 34 34 137 34 34

36

Figure 1: Composition of Privately-Produced Safe Debt (% of TotalPrivately-Produced Safe Debt)

0%  10%  20%  30%  40%  50%  60%  70%  80%  90%  100%  

1952Q1  

1955Q2  

1958Q3  

1961Q4  

1965Q1  

1968Q2  

1971Q3  

1974Q4  

1978Q1  

1981Q2  

1984Q3  

1987Q4  

1991Q1  

1994Q2  

1997Q3  

2000Q4  

2004Q1  

2007Q2  

2010Q3  

Deposits   Money-­‐like  debt  

MBS/ABS  Debt   Corporate  Bonds  and  Loans  

Other  LiabiliGes  

37

Figure 2: Holders of U.S. Treasury Securities (percent of total outstanding).

-­‐0.10  

0.00  

0.10  

0.20  

0.30  

0.40  

0.50  

0.60  

0.70  

1945  

1948  

1951  

1954  

1957  

1960  

1963  

1966  

1969  

1972  

1975  

1978  

1981  

1984  

1987  

1990  

1993  

1996  

1999  

2002  

2005  

2008  

2011  

US  Depository  Ins;tu;ons   Rest  of  the  World  Insurance  Companies   Mutual  Funds  

38

Figure 3: Fails by type.

0  

500,000  

1,000,000  

1,500,000  

2,000,000  

2,500,000  

3,000,000  

7/4/19

90  

7/4/19

92  

7/4/19

94  

7/4/19

96  

7/4/19

98  

7/4/20

00  

7/4/20

02  

7/4/20

04  

7/4/20

06  

7/4/20

08  

7/4/20

10  

7/4/20

12  

7/4/20

14  

$  Millions  

Panel  A:  Primary  Dealer  Treasury  Fails  

Total  Treasury  Receive   Total  Treasury  Deliver  

0  

50,000  

100,000  

150,000  

200,000  

250,000  

300,000  

7/4/1990  

7/4/1992  

7/4/1994  

7/4/1996  

7/4/1998  

7/4/2000  

7/4/2002  

7/4/2004  

7/4/2006  

7/4/2008  

7/4/2010  

7/4/2012  

7/4/2014  

$  Millions  

Panel  B:  Primary  Dealer  Agency  Fails  

Agency  Receive   Agency  Deliver  

0  

200,000  

400,000  

600,000  

800,000  

1,000,000  

1,200,000  

1,400,000  

7/4/1990  

7/4/1992  

7/4/1994  

7/4/1996  

7/4/1998  

7/4/2000  

7/4/2002  

7/4/2004  

7/4/2006  

7/4/2008  

7/4/2010  

7/4/2012  

7/4/2014  

$  Millions  

Panel  C:  Primary  Dealer  MBS  Fails  

MBS  Receive   MBS  Deliver  

39

Figure 4: DTCC Fails ($100 million).

0  

100  

200  

300  

400  

500  

600  

700  

9/19/2013  

10/19/2013  

11/19/2013  

12/19/2013  

1/19/2014  

2/19/2014  

3/19/2014  

4/19/2014  

5/19/2014  

6/19/2014  

7/19/2014  

8/19/2014  

9/19/2014  

$100  m

illions  

Treasury  Fails  ($100  mil.)   Agency  Fails  ($100  mil.)  

40

Figure 5: Difference between fail to receive and failure to deliver by type.

-­‐100,000  

-­‐50,000  

0  

50,000  

100,000  

150,000  

200,000  

250,000  

7/4/19

90  

7/4/19

92  

7/4/19

94  

7/4/19

96  

7/4/19

98  

7/4/20

00  

7/4/20

02  

7/4/20

04  

7/4/20

06  

7/4/20

08  

7/4/20

10  

7/4/20

12  

7/4/20

14  

$  Millions  

Panel  A:  U.S.  Primary  Dealers'  Repo  Fails:  US  Treasuries  Receive  minus  Deliver  

-­‐60,000  -­‐50,000  -­‐40,000  -­‐30,000  -­‐20,000  -­‐10,000  

0  10,000  20,000  30,000  40,000  

7/4/19

90  

7/4/19

92  

7/4/19

94  

7/4/19

96  

7/4/19

98  

7/4/20

00  

7/4/20

02  

7/4/20

04  

7/4/20

06  

7/4/20

08  

7/4/20

10  

7/4/20

12  

7/4/20

14  

$  Millions  

Panel  B:  U.S.  Primary  Dealers'  Repo  Fails:  Agencies  Receive  minus  Deliver  

-­‐200,000  

-­‐150,000  

-­‐100,000  

-­‐50,000  

0  

50,000  

100,000  

7/4/19

90  

7/4/19

92  

7/4/19

94  

7/4/19

96  

7/4/19

98  

7/4/20

00  

7/4/20

02  

7/4/20

04  

7/4/20

06  

7/4/20

08  

7/4/20

10  

7/4/20

12  

7/4/20

14  

$  Millions  

Panel  C:  U.S.  Primary  Dealers'  Repo  Fails:  MBS  Receive  minus  Deliver  

41

Figure 6: Panel A plots the fraction of Treasuries held to back notes andhence is informative about how aggressively banks were taking advantage ofnote issuance. Panel B plots the fraction of bonds on hand to loans anddiscounts. Panel C plots total Debt/GDP outstanding for the US and givesa sense of the total supply of government debt.

0%  

5%  

10%  

15%  

20%  

25%  

30%  

1874-­‐01-­‐01  

1875-­‐06-­‐01  

1876-­‐11-­‐01  

1878-­‐04-­‐01  

1879-­‐09-­‐01  

1881-­‐02-­‐01  

1882-­‐07-­‐01  

1883-­‐12-­‐01  

1885-­‐05-­‐01  

1886-­‐10-­‐01  

1888-­‐04-­‐01  

1889-­‐09-­‐01  

1891-­‐03-­‐01  

1892-­‐08-­‐01  

1894-­‐03-­‐01  

1895-­‐08-­‐01  

1897-­‐01-­‐01  

1898-­‐06-­‐01  

1899-­‐11-­‐01  

1901-­‐04-­‐01  

1902-­‐09-­‐01  

1904-­‐02-­‐01  

1905-­‐07-­‐01  

1906-­‐12-­‐01  

1908-­‐05-­‐01  

1909-­‐10-­‐01  

Treasuries  Held  to  Back  Notes  as  a  Percentage  of  Total  U.S.  Debt  Outstanding  

0.00%  

1.00%  

2.00%  

3.00%  

4.00%  

5.00%  

6.00%  

7.00%  

8.00%  

1/22/1870  

10/2/1871  

4/25/1873  

12/31/1874  

6/30/1876  

3/15/1878  

10/2/1879  

5/6/1881  

12/30/1882  

6/20/1884  

3/1/1886  

10/5/1887  

5/13/1889  

12/19/1890  

7/12/1892  

2/28/1894  

9/28/1895  

5/14/1897  

12/1/1898  

6/29/1900  

2/25/1902  

9/9/1903  

3/14/1905  

11/12/1906  

7/15/1908  

1/31/1910  

9/1/1911  

4/4/1913  

Ra1o  of  U.S.  Bonds  on  Hand  to  Loans  and  Discounts  

0.0  

5.0  

10.0  

15.0  

20.0  

25.0  

30.0  

35.0  

1870  1872  1874  1876  1878  1880  1882  1884  1886  1888  1890  1892  1894  1896  1898  1900  1902  1904  1906  1908  1910  1912  

US  Debt/GDP  

42

Figure 7: Original profit series as computed by Champ.

that the amount of tied-up capital would be zero. Presumably, the bank could earn

infinite profits by using the acquired notes to purchase additional government bonds

ad infinitum.6

Most certainly, the implied profit rates are, at times, far in excess of rates of return

on alternative uses of bank capital. Figure 5.2 shows my calculations of the profit

rates on issuing national bank notes using the formula suggested by Cagan. This

Figure 5.2: Cagan’s Profit Rate on National Bank Note Issuance, 1878–1913 (percent per annum)

250

200

150

100

50

0

1910190519001895189018851880

Source: Author’s calculations using Cagan’s profit rate formula, equation 5.1. The bond price datacomes from various issues of the Commercial and Financial Chronicle. The costs of note issue werecompiled from data presented in the Annual Report of the Secretary of the Treasury.

data portrayed in this figure uses more accurate representations of the costs of note

issue than those estimated by the Comptroller of the Currency. Rather than being

constant as in Cagan’s original formulation, the cost of note issue estimates presented

6Kuhlwein (1992) criticizes this view by claiming that dealers in government bonds may not havealways accepted national bank notes in payment.

5

43

Figure 8: Profit series, plotted in standard (top) and log scale (bottom).

0  

20  

40  

60  

80  

100  

120  

140  

0  

0.1  

0.2  

0.3  

0.4  

0.5  

0.6  

0.7  

0.8  

0.9  

1  

1875   1878   1881   1884   1887   1890   1893   1896   1899   1902   1905   1908   1911   1914  Recession   Profit  

1  

10  

100  

1000  

0  

0.1  

0.2  

0.3  

0.4  

0.5  

0.6  

0.7  

0.8  

0.9  

1  

1875   1878   1881   1884   1887   1890   1893   1896   1899   1902   1905   1908   1911   1914  Recession   Profit  

44

Figure 9: Ratio of notes to deposits. This figure plots the ratio of notesto deposits against US government debt to GDP. It shows that declines ingovernment supply of Treasuries are strongly associated with increases indeposits.

0

5

10

15

20

25

30

35

0

10

20

30

40

50

60

18

64

18

66

18

68

18

70

18

72

18

74

18

76

18

78

18

80

18

82

18

84

18

86

18

88

18

90

18

92

18

94

18

96

18

98

19

00

19

02

19

04

19

06

19

08

19

10

19

12

19

14

Ratio Notes/Deposits Debt / GDP

45

Figure 10: Safe asset share from Gorton Lewellen and Metrick 2012.

0.0%  

5.0%  

10.0%  

15.0%  

20.0%  

25.0%  

30.0%  

35.0%  

40.0%  

45.0%  

Government  Liabili7es   Financial  Liabili7es  

46

Appendices

Table 5: We report breakpoints for our fails data panel for the absolute valueof the difference between fails to receive and fails to deliver along with 95%confidence intervals. The methodology for finding breaks in panels followsBai (2010).

Panel A: All securities, |receive − deliver|Break Date Lower Bound Upper Bound

First Break 12-Sep-01 25-Jul-01 24-Oct-01Second Break 19-Aug-09 1-Jul-09 30-Sep-09Third Break 1-Feb-12 23-Nov-11 4-Apr-12

Panel B: Treasury+Agency, |receive − deliver|Break Date Lower Bound Upper Bound

First Break 6-Dec-00 31-May-00 6-Jun-01Second Break 24-Sep-08 25-Jun-08 17-Dec-08Third Break 11-Feb-09 17-Dec-08 1-Apr-09

Panel C: MBS, |receive − deliver|Break Date Lower Bound Upper Bound

First Break 19-Jun-02 1-May-02 31-Jul-02Second Break 19-Aug-09 1-Jul-09 30-Sep-09Third Break 1-Feb-12 9-Nov-11 18-Apr-12

47

Table 6: We run seemingly unrelated regressions of Treasury and Agencyfails to deliver.

∆ Fails Del ∆ Fails Del ∆ Fails Del ∆ Fails Del ∆ Fails Del ∆ Fails DelGC Repo-1m T-bill 6.727∗∗∗ 0.598 7.002∗∗∗ 0.552 7.207∗∗∗ 0.562

(5.43) (0.35) (5.58) (0.33) (5.72) (0.33)

L1.GC Repo-1m T-bill 2.118 0.770 2.448 0.607(1.69) (0.45) (1.93) (0.36)

L2.GC Repo-1m T-bill 2.417 0.406(1.92) (0.24)

GC Repo-1m T-bill x Break 1 12.97∗∗∗ 13.54∗∗∗ 12.96∗∗∗

(5.02) (5.11) (4.88)

L1.GC Repo-1m T-bill x Break 1 2.172 1.673(0.83) (0.62)

L2.GC Repo-1m T-bill x Break 1 -1.971(-0.75)

GC Repo-1m T-bill x Break 2 35.82∗∗∗ 40.87∗∗∗ 39.41∗∗∗

(6.69) (7.57) (7.32)

L1.GC Repo-1m T-bill x Break 2 27.47∗∗∗ 31.77∗∗∗

(5.42) (6.24)

L2.GC Repo-1m T-bill x Break 2 29.20∗∗∗

(5.81)

GC Repo-1m T-bill x Break 3 -0.421 -0.979 -0.152(-0.04) (-0.08) (-0.01)

L1.GC Repo-1m T-bill x Break 3 2.129 2.970(0.19) (0.25)

L2.GC Repo-1m T-bill x Break 3 3.371(0.29)

Break 1 (9/2001-9/2008) -6.809 -6.450 -4.782(-0.18) (-0.17) (-0.12)

Break 2 (9/2008-2/2009) 62.38 126.5 163.3(0.47) (0.96) (1.25)

Break 3 (2/2009) 10.22 9.742 6.906(0.24) (0.23) (0.16)

D. 1m T-Bill -11.20∗∗∗ -9.827∗∗∗ -7.917∗∗∗

(-5.52) (-4.77) (-3.81)

Constant 0.966 -10.30 1.273 -8.808 1.145 -6.793(0.06) (-0.42) (0.08) (-0.36) (0.07) (-0.28)

Observations 2398 2398 2386 2386 2374 2374R2 0.012 0.047 0.013 0.062 0.015 0.077

t statistics in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

48

Table 7: We run seemingly unrelated regressions of Treasury and Agencyfails to deliver.

∆ D-R ∆ D-R ∆ D-R ∆ D-R ∆ D-R ∆ D-RGC Repo-1m T-bill -0.0555 -0.0353 -0.0555 -0.0274 -0.0457 0.00581

(-0.25) (-0.12) (-0.25) (-0.09) (-0.20) (0.02)

L1.GC Repo-1m T-bill 0.0254 -0.0181 0.0472 -0.0198(0.11) (-0.06) (0.21) (-0.07)

L2.GC Repo-1m T-bill 0.185 0.186(0.82) (0.63)

GC Repo-1m T-bill x Break 1 0.118 0.327 0.307(0.27) (0.71) (0.66)

L1.GC Repo-1m T-bill x Break 1 0.606 0.726(1.34) (1.55)

L2.GC Repo-1m T-bill x Break 1 0.184(0.40)

GC Repo-1m T-bill x Break 2 0.351 -0.00556 -0.0348(0.38) (-0.01) (-0.04)

L1.GC Repo-1m T-bill x Break 2 -1.920∗ -1.933∗

(-2.19) (-2.18)

L2.GC Repo-1m T-bill x Break 2 -0.280(-0.32)

GC Repo-1m T-bill x Break 3 0.957 0.618 0.464(0.48) (0.30) (0.22)

L1.GC Repo-1m T-bill x Break 3 -0.840 -1.049(-0.43) (-0.52)

L2.GC Repo-1m T-bill x Break 3 -0.260(-0.13)

Break 1 (9/2001-9/2008) 89.87∗∗∗ 89.50∗∗∗ 89.39∗∗∗

(13.55) (13.47) (13.40)

Break 2 (9/2008-2/2009) 268.3∗∗∗ 263.4∗∗∗ 263.2∗∗∗

(11.88) (11.61) (11.56)

Break 3 (2/2009) 21.07∗∗ 21.40∗∗ 21.69∗∗

(2.92) (2.95) (2.97)

D. 1m T-Bill -0.373 -0.544 -0.555(-1.07) (-1.53) (-1.54)

Constant 74.67∗∗∗ 37.40∗∗∗ 74.91∗∗∗ 37.34∗∗∗ 75.11∗∗∗ 37.37∗∗∗

(24.98) (8.81) (24.95) (8.77) (24.89) (8.73)Observations 2398 2398 2386 2386 2374 2374R2 0.000 0.114 0.000 0.117 0.000 0.117

t statistics in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

49

Table 8: We run seemingly unrelated regressions of MBS fails to receive.∆ Fails Rec ∆ Fails Rec ∆ Fails Rec ∆ Fails Rec ∆ Fails Rec ∆ Fails Rec

GC Repo-1m T-bill 6.963∗∗∗ 6.113∗∗∗ 7.303∗∗∗ 6.124∗∗∗ 7.509∗∗∗ 6.501∗∗∗

(5.57) (3.59) (5.78) (3.60) (5.91) (3.76)

L1.GC Repo-1m T-bill 2.609∗ -1.192 2.951∗ -1.202(2.07) (-0.70) (2.31) (-0.70)

L2.GC Repo-1m T-bill 2.495∗ 2.200(1.96) (1.27)

GC Repo-1m T-bill x Break 1 5.348∗ 7.949∗∗ 7.532∗∗

(2.11) (3.08) (2.89)

L1.GC Repo-1m T-bill x Break 1 10.71∗∗∗ 11.75∗∗∗

(4.20) (4.50)

L2.GC Repo-1m T-bill x Break 1 1.635(0.63)

GC Repo-1m T-bill x Break 2 -26.33 -26.27 -26.60(-1.03) (-1.02) (-1.04)

L1.GC Repo-1m T-bill x Break 2 -1.043 -1.480(-0.04) (-0.06)

L2.GC Repo-1m T-bill x Break 2 11.23(0.44)

GC Repo-1m T-bill x Break 3 -0.995 -1.673 -0.807(-0.07) (-0.11) (-0.05)

L1.GC Repo-1m T-bill x Break 3 2.679 3.554(0.18) (0.22)

L2.GC Repo-1m T-bill x Break 3 -0.644(-0.04)

Break 1 (10/2002-9/2009) -19.14 -19.15 -19.33(-0.49) (-0.50) (-0.50)

Break 2 (9/2009-12/2011) 18.91 19.14 17.79(0.32) (0.33) (0.31)

Break 3 (12/2011) 11.41 12.42 10.48(0.20) (0.22) (0.18)

D. 1m T-Bill -12.98∗∗∗ -13.11∗∗∗ -12.90∗∗∗

(-6.32) (-6.37) (-6.23)

Constant 0.879 -9.285 1.104 -9.406 1.092 -8.803(0.05) (-0.39) (0.07) (-0.39) (0.06) (-0.37)

Observations 2398 2398 2386 2386 2374 2374R2 0.013 0.030 0.015 0.040 0.016 0.043

t statistics in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

50

Table 9: We run seemingly unrelated regressions of MBS fails to deliver.∆ Fails Del ∆ Fails Del ∆ Fails Del ∆ Fails Del ∆ Fails Del ∆ Fails Del

GC Repo-1m T-bill 6.727∗∗∗ 5.459∗∗ 7.002∗∗∗ 5.464∗∗ 7.207∗∗∗ 5.908∗∗∗

(5.43) (3.23) (5.58) (3.23) (5.72) (3.43)

L1.GC Repo-1m T-bill 2.118 -1.350 2.448 -1.358(1.69) (-0.80) (1.93) (-0.80)

L2.GC Repo-1m T-bill 2.417 2.540(1.92) (1.48)

GC Repo-1m T-bill x Break 1 5.776∗ 8.097∗∗ 7.629∗∗

(2.30) (3.15) (2.95)

L1.GC Repo-1m T-bill x Break 1 9.847∗∗∗ 10.68∗∗∗

(3.88) (4.11)

L2.GC Repo-1m T-bill x Break 1 0.505(0.20)

GC Repo-1m T-bill x Break 2 -24.01 -23.83 -24.22(-0.94) (-0.93) (-0.95)

L1.GC Repo-1m T-bill x Break 2 -4.492 -5.109(-0.18) (-0.20)

L2.GC Repo-1m T-bill x Break 2 15.98(0.63)

GC Repo-1m T-bill x Break 3 -1.485 -2.366 -1.309(-0.10) (-0.15) (-0.08)

L1.GC Repo-1m T-bill x Break 3 2.340 3.601(0.16) (0.23)

L2.GC Repo-1m T-bill x Break 3 -0.467(-0.03)

Break 1 (10/2002-9/2009) -17.26 -17.25 -17.50(-0.45) (-0.45) (-0.45)

Break 2 (9/2009-12/2011) 18.34 18.67 17.12(0.32) (0.32) (0.30)

Break 3 (12/2011) 9.075 10.96 8.271(0.16) (0.19) (0.14)

D. 1m T-Bill -11.10∗∗∗ -11.19∗∗∗ -11.06∗∗∗

(-5.44) (-5.47) (-5.37)

Constant 0.966 -7.621 1.273 -7.690 1.145 -7.168(0.06) (-0.32) (0.08) (-0.32) (0.07) (-0.30)

Observations 2398 2398 2386 2386 2374 2374R2 0.012 0.026 0.013 0.034 0.015 0.036

t statistics in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

51

Table 10: We run seemingly unrelated regressions of MBS fails to deliver.∆ D-R ∆ D-R ∆ D-R ∆ D-R ∆ D-R ∆ D-R

GC Repo-1m T-bill -0.0555 0.0792 -0.0555 0.0830 -0.0457 0.145(-0.25) (0.27) (-0.25) (0.28) (-0.20) (0.48)

L1.GC Repo-1m T-bill 0.0254 0.114 0.0472 0.118(0.11) (0.38) (0.21) (0.40)

L2.GC Repo-1m T-bill 0.185 0.362(0.82) (1.20)

GC Repo-1m T-bill x Break 1 -0.201 -0.225 -0.282(-0.46) (-0.50) (-0.62)

L1.GC Repo-1m T-bill x Break 1 -0.201 -0.220(-0.45) (-0.48)

L2.GC Repo-1m T-bill x Break 1 -0.419(-0.93)

GC Repo-1m T-bill x Break 2 1.635 1.674 1.620(0.37) (0.38) (0.36)

L1.GC Repo-1m T-bill x Break 2 -1.450 -1.549(-0.32) (-0.35)

L2.GC Repo-1m T-bill x Break 2 2.371(0.53)

GC Repo-1m T-bill x Break 3 0.778 0.545 0.436(0.31) (0.20) (0.16)

L1.GC Repo-1m T-bill x Break 3 0.302 0.247(0.12) (0.09)

L2.GC Repo-1m T-bill x Break 3 0.00159(0.00)

Break 1 (10/2002-9/2009) 81.37∗∗∗ 81.22∗∗∗ 81.10∗∗∗

(12.13) (12.06) (12.00)

Break 2 (9/2009-12/2011) 11.96 11.89 11.67(1.19) (1.17) (1.15)

Break 3 (12/2011) 0.403 0.629 0.843(0.04) (0.06) (0.08)

D. 1m T-Bill -0.495 -0.509 -0.532(-1.39) (-1.42) (-1.47)

Constant 74.67∗∗∗ 48.45∗∗∗ 74.91∗∗∗ 48.58∗∗∗ 75.11∗∗∗ 48.65∗∗∗

(24.98) (11.68) (24.95) (11.66) (24.89) (11.63)Observations 2398 2398 2386 2386 2374 2374R2 0.000 0.065 0.000 0.065 0.000 0.065

t statistics in parentheses∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

52